8
$\begingroup$

Note that $a>0$, thus I'm not sure if we can apply residues here. (For $a=0$ the integral doesn't converge).

$$\int_0^\infty \frac{dx}{(1+e^x)(a+x)}$$

Despite the simple expression under the integral, I didn't find it in G-R book.

I attempted the most straightforward way - geometric series - since $e^x \geq 1$.

$$\int_0^\infty \frac{dx}{(1+e^x)(a+x)}=\sum_{k=1}^\infty (-1)^{k+1} \int_0^\infty \frac{e^{-kx}dx}{a+x}$$

The integrals on the right can be expressed using incomplete gamma function:

$$\int_0^\infty \frac{e^{-kx}dx}{a+x}=\frac{e^{ka}}{k} \Gamma(0,ka)$$

We obtain:

$$\int_0^\infty \frac{dx}{(1+e^x)(a+x)}=\sum_{k=1}^\infty (-1)^{k+1} \frac{e^{ka}}{k} \Gamma(0,ka)$$

This is correct, but not very useful, since each incomplete gamma just disguises an integral. It can be computed independently by:

$$\Gamma (0,t)=\cfrac{\exp(-t)}{t+1-\cfrac{1}{t+3-\cfrac{4}{t+5-\cfrac{9}{t+7-\cdots}}}}$$

Thus I suppose we can express the integral as:

$$\int_0^\infty \frac{dx}{(1+e^x)(a+x)}=\sum_{k=1}^\infty \frac{(-1)^{k+1}}{k} \cfrac{1}{ka+1-\cfrac{1}{ka+3-\cfrac{4}{ka+5-\cfrac{9}{ka+7-\cdots}}}}$$

Amusing, but again, not very useful.

Does this integral have a closed form? If so, what is it and how do we find it?

Edit

Another interesting expression for incomplete gamma from DLMF:

$$\Gamma (0,t)=e^{-t} \sum_{n=0}^\infty \frac{L_n (t)}{n+1}$$

Here $L_n (t)$ are Laguerre polynomials. This series converges extremely slowly (and oscillates), so it seem quite useless for computation, however gives a pretty expression for the integral:

$$\int_0^\infty \frac{dx}{(1+e^x)(a+x)}=\sum_{k=1}^\infty \frac{(-1)^{k+1}}{k} \sum_{n=0}^\infty \frac{L_n (ka)}{n+1} $$

Note, that the continued fraction above converges much faster than the series.

$\endgroup$
3
  • $\begingroup$ In the cause of a proof, I found a solution of the integral: math.stackexchange.com/questions/3433267/… $\endgroup$ – stocha Nov 23 '19 at 16:52
  • 1
    $\begingroup$ @stocha, very interesting, thank you $\endgroup$ – Yuriy S Nov 23 '19 at 18:53
  • $\begingroup$ Thanks, your contribution helped me with the proof! $\endgroup$ – stocha Nov 23 '19 at 22:18
4
$\begingroup$

Since: $$\mathcal{L}\left(\frac{1}{1+e^x}\right) = \frac{1}{2}\left[\psi\left(\frac{s+2}{2}\right)-\psi\left(\frac{s+1}{2}\right)\right],\qquad \mathcal{L}^{-1}\left(\frac{1}{a+x}\right)=e^{-as}$$ the original integral equals: $$ \int_{0}^{+\infty}\frac{1}{2}\left[\psi\left(\frac{s+2}{2}\right)-\psi\left(\frac{s+1}{2}\right)\right]e^{-as}\,ds=\frac{1}{2}\int_{0}^{+\infty}\int_{\frac{s+1}{2}}^{\frac{s+2}{2}}\sum_{n\geq 0}\frac{e^{-as}}{(n+u)^2}\,du\,ds $$ and by integration by parts, the LHS just depends on: $$ \int_{0}^{+\infty}\left[\log\Gamma\left(\frac{s+2}{2}\right)-\log\Gamma\left(\frac{s+1}{2}\right)\right]e^{-as}\,ds$$ that can be further expanded by exploiting the Weierstrass product for the $\Gamma$ function, leading to a fast-convergent series, whose first term is $-\frac{\gamma}{2a}$.

$\endgroup$
4
  • $\begingroup$ What is $\psi$? $\endgroup$ – GFauxPas Jun 10 '16 at 14:16
  • 1
    $\begingroup$ @GFauxPas, digamma function, I presume $\endgroup$ – Yuriy S Jun 10 '16 at 14:16
  • $\begingroup$ Do I understand correctly that Laplace transform in this case means just using $$\frac{1}{a+x}=\int_0^\infty e^{-(a+x)t}dt$$, reversing order of integration and then solving $$\int_0^\infty \frac{e^{-tx} dx}{1+e^x}$$? $\endgroup$ – Yuriy S Jun 10 '16 at 14:30
  • 1
    $\begingroup$ @YuriyS: absolutely right. $\endgroup$ – Jack D'Aurizio Jun 10 '16 at 14:31
0
$\begingroup$

Just some further thoughts, which neatly connect different methods of transforming the integral.

Using Jack D'Aurizio's answer, we work with the integral:

$$I(a)=\int_{0}^{+\infty}\frac{1}{2}\left[\psi\left(\frac{s+2}{2}\right)-\psi\left(\frac{s+1}{2}\right)\right]e^{-as}\,ds= \\ = \sum_{n=0}^\infty (-1)^n \int_{0}^{+\infty} \frac{e^{-a s} ds}{n+s+1}$$

This is actually the same series we get by expanding the original integrand in geometric series:

$$I(a)=\sum_{n=0}^\infty (-1)^n \int_0^\infty \frac{e^{-(n+1)x}dx}{x+a}$$

However, the first form of the series might be better for acceleration.

Using Euler's transform:

$$\sum_{k=0}^n (-1)^k \binom{n}{k} \frac{1}{n-k+s+1}=(-1)^n \mathrm B (n+1,s+1)$$

We have:

$$\sum_{n=0}^\infty \frac{(-1)^n}{n+s+1}=\sum_{n=0}^\infty \frac{\mathrm B (n+1,s+1)}{2^{n+1}}$$

Which gives us:

$$I(a)=\sum_{n=0}^\infty \frac{1}{2^{n+1}}\int_{0}^{+\infty} \mathrm B (n+1,s+1)e^{-a s} ds \tag{1}$$

Using the integral definition for the Beta function, we have:

$$I(a)=\sum_{n=0}^\infty \frac{1}{2^{n+1}}\int_{0}^{+\infty} \int_0^1 t^n (1-t)^s e^{-a s} dt ds $$

Summation gives us:

$$I(a)= \int_{0}^{+\infty} \int_0^1 \frac{(1-t)^s}{2-t} e^{-a s} dt ds $$

Integrating w.r.t. $s$:

$$I(a)= \int_0^1 \frac{dt}{(2-t)(a-\log(1-t))} = \int_0^1 \frac{dt}{(1+t)(a-\log t)}$$

Which, funny enough, brings us back to the original integral.


Let's get back to (1). The Beta functions and the exponential are both very nice functions, which fall to $0$ fast.

$$I(a)=\sum_{n=0}^\infty \frac{1}{2^{n+1}}\int_{0}^{+\infty} \mathrm B (n+1,s+1)e^{-a s} ds$$

enter image description here

Finding a suitable approximation for the Beta functions (conventional asymptotics are nor very useful in this case), we can get a fast converging series for $I(a)$. There's probably a good approximation as a sum of two or three exponential terms.


The integration by parts advice from Jack is sound as well, but I haven't worked out the resulting series yet.

$$I(a)=\frac{\log \pi}{2} + a \int_0^\infty \left[\log\Gamma\left(\frac{s+2}{2}\right)-\log\Gamma\left(\frac{s+1}{2}\right)\right]e^{-as} ds$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.